The Silencio voice network, right now:
2,000,000 contributors
180 countries
150 distinct languages
5,000 hours of new audio added every day
What this scale buys, technically: enough variation in voices, accents, devices, and environments that a model trained on it generalises to the real world, not to a studio. Models trained on 50 voices fail outside the lab. Models trained on a million voices don't.
Every recording adds one more degree of "the model has heard someone like this before."